Too little content

to much shallow

Good course

too basic

just ok

Materials were good for someone who has taken university courses on convolutional networks, but labs were extremely poorly done. Final lab of the course was missing sections for the data generator flow method calls, and augmentation wasn't even tested for. Marker could be improved and provided code can have better sections and maybe an explaining markdown at the top rather than going back and forth. I also noticed that accuracy changed from logs.get('acc') to logs.get('accuracy') which seems to be a tensorflow version issue. I feel overall like the course has been abandoned.

If you have taken Andrew's courses in ML or deep learning, you will be disappointed. The amount of content in the videos and exercises is shrunk down by 75% per week. I think a much better job could have been done of structuring the course, and creating meaningful exercises. The instructor does an OK job of showing you how to use TF, but he doesn't always explain things very clearly, and doesn't always have an accurate understanding of how ML or deep learning works.

Content is really useful, but the assignment is really really bad and not user friendly(actually it drives me crazy). For example, instruction is not clear, parameter is outdated(still use 'acc' for accuracy?), assignment cannot be graded not because of modeling. These inconvenience obscure of the importance of learning CNN in TF. For this reason I don't think this course worth more than 3 stars.

This course could have covered many more topics in detail, like visualizing individual layers, performing style transfer, saving and loading models, etc. All these were skipped and weeks were wasted on a simple extension of a small concept (image augmentation and multi-class learning) which anyone who glanced at the Keras API could have learnt. I am disappointed at this course frankly.

Really a bad course. Most of the materials can be found online for free on TensorFlow official documentations. Many practices are outdated. Problems with the coding assignment are a nightmare. There is no supervisor to answer many common questions. The code grader checks for very particular things and instructions were not clear at all. In general, this is a pretty bad course.

Course Material not detailed enough and expected more from it. It does not contain enough variety in exercises and lacks a lot of concepts.

Anyone with good learning (and "overfitting") can complete 1 course in a day.

Final assignment lacked a lot of use case description and it did not even tell us anything about the data or recommended parameters for training.

There was a lot of repetition with respect to the previous course, making this one feel a bit like filler. The exercises were poorly designed: they used functionality that was not taught during the course, were badly explained and often had weird passing conditions, to the point that I felt I was fighting with the grader instead of trying to improve my code.

Terrible information flow. Submissions are not connected to any real world knowledge, neither to Machine Learning nor to tensorflow. Information provided assumes that you have some knowledge of ML, so no ML information given, but if you assume so, information about tensorflow is extremely basic and not learning progress not checked properly.

Far to simple. Significant concepts were glossed over and the exercises were mainly copy and past from the examples. Lessons that covered a "week" took < 1 hour with a couple minor points learned. Don't recommend if you want to really know how CNNs work.

The content could have been covered in 15 min. Moreover, I thought you guys will teach tensorflow low level API and estimators etc. atleast in course 2. Also, what happened to the graded assignments ? I finished this course in 40 min.

The course simply does not cover much information. The whole course could be compacted into a decent one hour lecture. Andrew Ng has some great courses on machine learning but I don't believe this to be one of them.

I was wasting my time with small coding obstacles for setting up the data (missing hints ?!) instead of getting teached on how to implement models in various ways and how to use regularization for real...

Again very superficial. Yet, some assignments required tweaks that were never teached (i.e. which optimizer to use when, when to use "sparse_categorical_crossentropy", etc.), which was frustrating.

The course material is good and the lectures are great but rating it low to bring attention to the assignments which were old and incomplete. They really need to be updated. Specially week 4.

In my opinion there is to little background explanation. There were errors in Programming Assignment

in week 3 and 4. I had to look in forum discussion so I could complete the exercise

weeks from 1-3 were perfect.

But the programming assignment on week 4 needs to be fixed. Please add hints and examples, otherwise it is just a headache

No enough time spend in the actual code which limits the learning on the understanding of the concepts with implementation. Doubt how useful this is

Last exercise is so bad. It didn't even discuss about flow function, why it's need 26 last layer? isn't it should be only 3?

Absolutely awful grader. You spend most of the time figuring out errors and intentions of the grader writer.

Very superficial look at deep learning. A lot of the programming assignments had little to no context.